subtopic

K-Nearest Neighbours (classifier | regressor)

Fit KNN with k = 1…50 on scaled features; plot validation score vs k. Concepts: Similarity-based prediction; distance metrics; choice of k; curse of dimensionality Project A: k ≈ 15–30 smooths noisy neighbours; usually below logistic regression here. Project B: KNN regressor on (hour, weekday, temperature): decent but cannot extrapolate beyond seen temperatures.